Sound wave W-VSP reverse time migration noise suppression method based on common detection point imaging

Abstract

The invention discloses a sound wave W-VSP reverse time migration noise suppression method based on common detection point imaging. The method comprises the steps that 1, finite difference sound wave fluctuation forward modeling is achieved through a PIRNN network; step 2, extracting a common detection point gather: extracting the seismic data as signals with the same detection points but different seismic sources to obtain the common detection point gather; step 3, for each detection point, obtaining a seismic source wave field and a detection point wave field by using the common detection point gather; step 4, through cross correlation of a seismic source wave field and a detection point wave field, imaging is carried out by using a detection point normalization cross correlation imaging condition; step 5, Laplace filtering is carried out on single detection point imaging; and step 6, superposing the single shot imaging results of the detection points to obtain a final common detection point reverse time migration imaging image. The invention provides a simple and efficient RTM noise suppression method, which utilizes information imaging of a full wave field and realizes suppression of arc marking noise.

Description

Technical Field

[0001] The invention belongs to the field of seismic data processing, and in particular relates to a reverse time migration imaging technology. Background Art

[0002] The RTM imaging method was originally proposed by scholars such as Baysal and Whitmore, opening a new chapter in the field of seismic imaging. Over the past few decades, RTM technology has developed significantly, especially in seismic imaging in complex geological areas, showing its outstanding advantages. Because VSP data has higher resolution and signal-to-noise ratio, compared with surface seismic data, it has significant advantages in the identification of complex underground structures, especially high-steep structures, which has attracted scholars to study the reverse time migration imaging method of VSP. However, although VSP's RTM technology has made significant progress in theory and application, low-frequency noise is still one of the key issues affecting its imaging accuracy. When using full waveform data for RTM, low-frequency noise is often unavoidable, and these noises have a great impact on the imaging quality. In order to suppress these noises, researchers have proposed a variety of noise suppression methods. At present, noise suppression methods for RTM are mainly divided into three categories:

[0003] The first category is to modify the two-way wave equation and improve the accuracy of wave field numerical simulation. This type of method mainly modifies the two-way wave equation or smoothes the velocity field to eliminate the main noise source during the wave field propagation process, especially to suppress the reflection noise in the interaction between the wave field and the boundary. These methods usually rely on the improvement of numerical algorithms to suppress artifacts and noise caused by wave propagation during the calculation process. In order to solve the problems of numerical dispersion and low-frequency noise, scholars have proposed a finite difference method based on LS optimization. This method effectively suppresses the errors caused by dispersion and improves the accuracy of wave field reconstruction. Especially under complex geological structures, the optimized FD method can significantly reduce high-frequency noise and dispersion errors by fine-tuning the length of the differential operator.

[0004] The second category is to achieve noise suppression by transforming imaging conditions. The following are several typical imaging condition improvement and reconstruction methods:

[0005] 1) Use the Poynting vector angle filter. For example, the Poynting vector-based angle filter aims to improve the imaging quality by filtering out large-angle noise. The Poynting vector is used to describe the propagation direction of wave energy. By angularly screening the Poynting vector in the wave field, the noise whose propagation direction does not meet the expected direction can be effectively removed. This method mainly solves the noise problem caused by excessive propagation angles in RTM imaging. In complex geological environments, RTM may produce some directional deviation artifacts, especially in high-angle reflection or transmission bands. By using the Poynting vector angle filter, the impact of these directional noises on the imaging results can be reduced, further improving the imaging signal-to-noise ratio.

[0006] 2) Implicit wavefield decomposition imaging conditions, such as the imaging conditions proposed by scholars to perform implicit wavefield decomposition in the wavenumber domain along the rising and falling directions. This method removes artifacts caused by reflection and transmission waves by decomposing the wavefield into components that propagate upward and downward, and cross-correlating them with the detection point wavefield respectively. By separating the upper and lower wavefields, the imaging conditions can retain only the reflected waves and suppress the influence of the transmitted waves or other irrelevant waves. This method can effectively deal with imaging problems in high-steep structures and complex geological environments, especially in the VSP data processing process. With this method, RTM can improve imaging accuracy and reduce the influence of interface artifacts under complex geological conditions.

[0007] 3) Explicit wavefield decomposition. In order to solve the problem of low-frequency noise, scholars proposed the imaging conditions of explicit wavefield decomposition and suggested selecting the source downlink wavefield and the detection point uplink wavefield to reduce noise interference. The core idea of ​​explicit wavefield decomposition is to avoid the interference of artifacts by clearly distinguishing the uplink and downlink wavefields, especially in environments with multiple reflections and complex wave propagation paths. The eight-way explicit wavefield decomposition method was further proposed. By explicitly decomposing the wavefield in multiple directions, the influence of low-frequency noise was further reduced. The eight-way decomposition technology can more comprehensively capture all angles of wavefield propagation, thereby reducing the noise problem caused by inaccurate imaging conditions.

[0008] The third category is post-imaging processing filtering and denoising, for example, applying Laplace operator filtering and diffusion filtering to remove low-frequency noise.

[0009] Compared with surface seismic, vertical seismic profile (VSP) can provide richer wave field information and higher resolution, high-quality seismic data. However, due to the particularity of the VSP observation system, it leads to the problem of small coverage and serious uneven distribution. Under the traditional normalized imaging conditions, the noise around the well is further amplified, especially under the low coverage conditions of W-VSP, it is difficult to suppress the arcing noise around the well during stacking. Summary of the invention

[0010] The purpose of the present invention is to overcome the shortcomings of the prior art and provide an acoustic W-VSP reverse time migration noise suppression method based on common detection point imaging. The present invention utilizes full wave field information and simple and efficient noise suppression technology to improve the signal-to-noise ratio within the imaging range and suppress low-frequency noise and arcing noise.

[0011] The object of the present invention is achieved through the following technical scheme: an acoustic wave W-VSP reverse time migration noise suppression method based on common detection point imaging, comprising the following steps:

[0012] The acoustic wave W-VSP reverse time migration noise suppression method based on common detection point imaging is characterized by comprising the following steps:

[0013] Step 1: Use the PIRNN network to implement finite difference acoustic wave forward modeling; embed the wave equation into the RNN network as a RNN basic unit to form a PIRNN network; the input of the RNN is the wave field data of the previous moment, and the network generates the wave field record of the current moment through the physical constraints of the finite difference of the wave equation;

[0014] Step 2: Extract common detection point gathers: After the observation system is deployed, the seismic wavelet is excited at each source in turn, and the signal is recorded at all detection points. The collection of these signals is the seismic data; the seismic data is extracted into signals with the same detection point but different sources to obtain common detection point gathers;

[0015] Step 3: For each detection point, use the common detection point gather to obtain the source wave field and the detection point wave field; place the common detection point gather at the corresponding source as excitation, and use the PIRNN network to reverse time-extend to obtain the source wave field; at the same time, stimulate the seismic wavelet at the detection point, and use the PIRNN network to forward extrapolate to obtain the detection point wave field;

[0016] Step 4: Through the cross-correlation of the source wave field and the detection point wave field, imaging is performed using the normalized cross-correlation imaging condition of the detection point;

[0017] Step 5: Perform Laplace filtering on the single detection point imaging;

[0018] Step 6: Superimpose the single-shot imaging results of each detection point to obtain the final common detection point reverse time migration imaging image.

[0019] In step 4, the imaging condition is expressed as:

[0020]

[0021] I(x,z) represents the imaging result, S(x,z,t max -t) represents the source wavefield, and R(x,z,t) represents the detection point wavefield.

[0022] The beneficial effects of the present invention are as follows: the present invention significantly improves the imaging quality of VSP reverse time migration imaging, and by redistributing noise energy, transfers strong noise to the vicinity of the surface source, thereby significantly improving the signal-to-noise ratio in the imaging area around the well. At the same time, compared with the traditional VSP common shot point RTM, the present invention can effectively expand the imaging area, especially for areas far away from the well, with higher imaging accuracy. In addition, the present invention utilizes the imaging condition of normalization of the detection point to effectively suppress low-frequency noise and arc noise, reduce imaging artifacts around the well, and improve the reliability of imaging. Through experimental verification on complex structures such as the Marmousi model, the method shows good adaptability and robustness. The present invention utilizes full wavefield information and simple and efficient noise suppression technology, which is significantly superior to traditional methods in imaging resolution and imaging quality, and can clearly identify complex geological structures, providing new technical means and research ideas for resource exploration, risk assessment and underground structure analysis under complex geological conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a flow chart of the reverse time migration noise suppression method of the present invention;

[0024] Figure 2 It is the RNN unit structure;

[0025] Figure 3 The process of realizing wave field forward modeling for RNN;

[0026] Figure 4 This is the CRRTM imaging principle;

[0027] Figure 5 is the horizontal velocity model;

[0028] Figure 6 The imaging results of the flat layer model, (a) is RTM, (b) is CRRTM;

[0029] Figure 7 This is the local model of Marmousi, where (a) the well is located at 3 km and (b) the well is located at 7 km;

[0030] Figure 8 Imaging results of different methods for marmousi. DETAILED DESCRIPTION

[0031] The present invention proposes a common detection point based reverse time migration (CRRTM) imaging method. Based on the second-order constant density acoustic wave equation, a PIRNN network is constructed for forward calculation of the wave field as the basis for imaging. The seismic data recorded by the common detection point is propagated backward at the corresponding source position to obtain the source wave field. The standard Ricker wavelet is excited at the detection point position to generate the detection point wave field. Subsequently, through the cross-correlation of the source wave field and the detection point wave field, combined with the imaging conditions of the detection point normalization, a high-resolution imaging result is obtained. The preliminary imaging results are subjected to post-processing operations such as Laplace filtering to further suppress low-frequency noise and artifacts and optimize the imaging quality. Finally, the single-shot imaging results of each detection point are superimposed to finally form a common detection point reverse time migration imaging image. The technical scheme of the present invention is further illustrated below in conjunction with the accompanying drawings.

[0032] like Figure 1 As shown, a method for suppressing acoustic wave W-VSP (Walkaway vertical seismic profiling) reverse time migration noise based on common detection point imaging of the present invention comprises the following steps:

[0033] Step 1: Use the PIRNN network to implement finite difference acoustic wave forward modeling; use the RNN network to implement staggered grid numerical simulation of acoustic wave equations. Recurrent Neural Networks (RNN) is an artificial neural network that specializes in processing sequence data. Its key feature is to use hidden states to store historical information so that the output can be affected by the input at the previous moment. RNN can be divided into two categories: finite impulse response network (FIR) and infinite impulse response network (IIR), both of which are suitable for dynamic time behavior modeling. This memory ability enables RNN to perform well in time series signal processing tasks such as text and audio. In seismic forward modeling, the wave equation describes the dynamic changes of the wave field in time and space, and is usually solved by the time step method, that is, the wave field distribution at the next moment is derived from the wave field state at the previous moment. This gradual update method is highly consistent with the mechanism of RNN processing time series data. In the present invention, the wave equation is embedded in the RNN network as a basic unit of the RNN, which is to embed physical information into the RNN network, so the PIRNN (PI is physics-informed, meaning physical information) is used in this article. By embedding the finite difference time-stepping method of the wave equation into the RNN unit, a new seismic forward modeling framework can be constructed. In this framework, the input of the RNN is the wave field data of the previous moment. The network generates the wave field record of the current moment through the physical constraints of the finite difference of the wave equation, and passes the result to the next moment to continue the calculation. This step-by-step iterative method can efficiently simulate the spatiotemporal evolution of the wave field. This combination not only improves the flexibility of seismic forward modeling, but also provides a new direction for optimizing numerical simulation using neural networks.

[0034] The staggered grid finite difference RNN operator network structure used in the present invention is described: This embodiment uses the first-order pressure-velocity fluctuation equation group of acoustic waves as an example to implement common detection point reverse time migration imaging:

[0035]

[0036] The above equation is numerically solved using staggered grid finite difference approximate derivatives and embedded in a recurrent neural network. Figure 2The basic unit of the finite difference RNN operator is shown. The input is the velocity parameter, the source position and the wave field information of the previous moment, and the output is the wave field information of the next moment. Both forward extrapolation and reverse time extension are completed using this model. The difference is that forward extrapolation directly inputs the excitation signal into the data model; while reverse time extension requires the excitation signal to be reversed along the time axis before input, and then reversed along the time axis after obtaining the final full wave field. In the present invention, the wave equation is embedded into the RNN network as a basic unit of the RNN network, which is to embed physical information into the RNN network. Therefore, the PIRNN (PI stands for physics-informed) used in this article is called.

[0037] Where p is the particle pressure, p x and p z are the pressure components in the x and z directions, w(t) is the excitation signal added to the source (the Ricker wavelet is used in this invention), v x ,v z is the particle velocity in the x and z directions, v is the speed of sound, i.e. the input imaging velocity model, d(x) and d(z) represent the boundary absorption coefficients, so as to avoid reflection at the model boundary interfering with the effective signal. The PML (Perfectly Matched Layer) absorption coefficient selected by the present invention is expressed as follows:

[0038]

[0039] Where L represents the PML boundary thickness, x represents the distance between the current position of the particle (inside the PML layer) and the inner boundary of the PML, and R represents the absorption coefficient, which is usually a constant.

[0040] Step 2, extract the common detection point gather: After the observation system is deployed (assuming there are n sources, m detection points, and a detector is deployed at each detection point), excite the seismic wavelet at each source in turn, and record the signal at all detection points. A total of n*m signals will be obtained. The collection of these signals is the seismic data, also commonly referred to as observation records; extract the seismic data into signals with the same detection points but different sources to obtain the common detection point gather.

[0041] Step 3: For each detection point, use the common detection point gather to obtain the source wave field and the detection point wave field; place the common detection point gather at the corresponding source as excitation, and use the PIRNN network to reverse time-extend to obtain the source wave field; at the same time, stimulate the seismic wavelet at the detection point, and use the PIRNN network to forward extrapolate to obtain the detection point wave field;

[0042] Step 4: Through the cross-correlation of the source wave field and the detection point wave field, imaging is performed using the normalized cross-correlation imaging condition of the detection point;

[0043] Step 5: Perform Laplace filtering on the single detection point imaging to suppress low-frequency noise and artifacts and optimize the imaging quality;

[0044] Step 6: Superimpose the single-shot imaging results of each detection point to obtain the final common detection point reverse time migration imaging image.

[0045] Figure 4 The imaging principle of the common detection point method is demonstrated, and the maximum time recorded by the detection point is t max The seismic wavelet is excited at the source, propagates to the reflection point after t1, and then propagates to the detector after t2. The reflected signal is recorded at time tt1+t2. The seismic record received by the detector point will be moved to the source for reverse propagation. According to the Fermat principle, the seismic signal propagating backward from the source point S also propagates along the minimum path. The propagation after time t1 is t max -t2 arrives at the reflection point. At the same time, the seismic wavelet excited by the detection point also arrives at the reflection point at t2 according to the Fermat principle. Therefore, the source wavefield and the detection point wavefield at the reflection point meet the imaging conditions. In step 4, CRRTM uses the detection point wavefield normalized imaging conditions for imaging, and its imaging conditions are expressed as:

[0046]

[0047] I(x,z) represents the imaging result, S(x,z,t max -t) represents the source wavefield, and R(x,z,t) represents the detection point wavefield.

[0048] For the traditional source normalized cross-correlation imaging condition, a large amount of low-frequency noise will accumulate around the VSP well logging, seriously polluting the imaging results. However, the common detection point imaging (CRRTM) of the present invention utilizes the energy relationship between noise and effective imaging signals, records observations at source excitation, and uses detection point wave field normalization to effectively improve the signal-to-noise ratio of wellbore imaging and expand the illumination range.

[0049] Effectiveness of the CRRTM method: In numerical calculations, the stability of the solution is related to whether non-physical oscillations or divergence will occur. If the time step and space step do not meet the stability conditions, the calculation results may deviate from the true solution or even diverge. In general, the stability condition of staggered grid finite difference can be described as:

[0050]

[0051] Where c i is the absolute value of the discretized difference weight, which needs to be calculated according to the difference order, defined as Some commonly used coefficients are as follows:

[0052]

[0053] Building under stability constraints Figure 5 The simple flat-layer reflection model shown has a spatial grid number of 800*800, a velocity between 3000 and 4000 m / s, spatial intervals Δz and Δx of 5 m, a time interval of 0.5 ms, a Ricker wavelet frequency of 30 Hz, 20 PML absorption boundaries, a staggered grid finite difference order of 20, seismic sources uniformly distributed every 20 m at z=0 m, and detectors uniformly distributed every 20 m at x=2000 m.

[0054] Figure 6 The imaging results of the RTM and CRRTM methods are shown. It can be seen that both methods can image the three reflection surfaces, but the difference is that the traditional method accumulates strong noise around the well, and the imaging axis is not clear enough. However, in CRRTM imaging, the imaging axis is clear and clean, the noise around the well is significantly suppressed, and the imaging range is also expanded to a certain extent, and the imaging quality is significantly improved. The experimental results prove the effectiveness of CRRTM on the VSP system, and also reflect the prospects and value of further application of this method in the future.

[0055] Adaptability to complex models: The CRRTM method has achieved good performance on simple models. We consider evaluating the adaptability of this method in more complex models. We resampled the standard 1.25m*1.25m marmousi longitudinal wave velocity model to a spatial interval of dz=5, dx=5, with a velocity distribution of 800-4700m / s, a time interval of 0.5ms, a Ricker wavelet frequency of 40hz, a finite difference order of 20, and a source uniformly distributed every 40m at z=0m. The detection points are arranged at intervals of 20m in different wells. In order to test the adaptability of this method in complex models, we took two more representative areas in the marmousi model for imaging. Figure 7 (a) is the model of the VSP well located at 3 km, which mainly tests the imaging effect under a simple flat structure. Figure 7 (b) is the model with the well logging location at 7 km, which is characterized by a relatively large formation dip and a more complex geological structure.

[0056] Figure 8The performance of RTM and CRRTM on two marmousi local models is shown, (a) RTM (model 1), (b) CRRTM ((model 1), (c) RTM ((model 2), and (d) CRRTM ((model 2). The white dotted boxes mark several areas with obvious contrast. It can be seen that the traditional RTM method accumulates a large amount of artifact noise around the logging, which seriously pollutes the effective signal, while the CRRTM method disperses the noise energy at the edge of the image and obtains a continuous and clear imaging axis within the imaging range.

[0057] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A method for suppressing acoustic W-VSP reverse time migration noise based on common detection point imaging, characterized in that: The following steps are involved: Step 1: Use the PIRNN network to implement finite difference acoustic wave forward modeling; embed the wave equation into the RNN network as a RNN basic unit to form a PIRNN network; the input of the RNN is the wave field data of the previous moment, and the network generates the wave field record of the current moment through the physical constraints of the finite difference of the wave equation; Step 2: Extract common detection point gathers: After the observation system is deployed, the seismic wavelet is excited at each source in turn, and the signal is recorded at all detection points. The collection of these signals is the seismic data; the seismic data is extracted into signals with the same detection point but different sources to obtain common detection point gathers; Step 3: For each detection point, use the common detection point gather to obtain the source wave field and the detection point wave field; The common detection point gathers are placed at the corresponding source as excitation, and the source wave field is obtained by reverse time extension using the PIRNN network; at the same time, seismic wavelets are excited at the detection point, and the detection point wave field is obtained by forward extrapolation using the PIRNN network; Step 4: Through the cross-correlation of the source wave field and the detection point wave field, imaging is performed using the normalized cross-correlation imaging condition of the detection point; Step 5: Perform Laplace filtering on the single detection point imaging; Step 6: Superimpose the single-shot imaging results of each detection point to obtain the final common detection point reverse time migration imaging image.

2. The acoustic wave W-VSP reverse time migration noise suppression method based on common detection point imaging according to claim 1 is characterized in that: In step 4, the imaging condition is expressed as: I(x,z) represents the imaging result, S(x,z,t max -t) represents the source wavefield, and R(x,z,t) represents the detection point wavefield.

Images